Handbook of Global Analysis - Read Online

Book Preview

Handbook of Global Analysis

You've reached the end of this preview. Sign up to read more!
Page 1 of 1

Handbook of Global Analysis

Demeter Krupka

David Saunders

Palacky University, Olomouc, Czech Republic

Table of Contents

Cover image

Title page



Chapter 1: Global aspects of Finsler geometry

1 Finsler metrics and connections

2 Geodesics in Finsler manifolds

3 Comparison theorems: Cartan—Hadamard theorem, Bonnet—Myers theorem, Laplacian and volume comparison

4 Rigidity theorems: Finsler manifolds of scalar curvature and locally symmetric Finsler metrics

5 Closed geodesics on Finsler manifolds, sphere theorem and the Gauss—Bonnet formula

Chapter 2: Morse theory and nonlinear differential equations

1 Introduction

2 Global theory

3 Local theory

4 Applications

5 Strongly indefinite Morse theory

6 Strongly indefinite variational problems

Chapter 3: Index theory

1 Introduction and some history

2 Fredholm operators – theory and examples

3 The space of Fredholm operators and K-theory

4 Elliptic operators and Sobolev spaces

5 The Atiyah-Singer Index Theorem

6 Generalizations

Chapter 4: Partial differential equations on closed and open manifolds

1 Introduction

2 Sobolev spaces

3 Non-linear Sobolev structures

4 Self-adjoint linear differential operators on manifolds and their spectral theory

5 The spectral value zero

6 The heat equation, the heat kernel and the heat flow

7 The wave equation, its Hamiltonian approach and completeness

8 Index theory on open manifolds

9 The continuity method for non-linear PDEs on open manifolds

10 Teichmüller theory

11 Harmonic maps

12 Non-linear field theories

13 Gauge theory

14 Fluid dynamics

15 The Ricci flow

Chapter 5: The spectral geometry of operators of Dirac and Laplace type

1 Introduction

2 The geometry of operators of Laplace and Dirac type

3 Heat trace asymptotics for closed manifolds

4 Hearing the shape of a drum

5 Heat trace asymptotics of manifolds with boundary

6 Heat trace asymptotics and index theory

7 Heat content asymptotics

8 Heat content with source terms

9 Time dependent phenomena

10 Spectral boundary conditions

11 Operators which are not of Laplace type

12 The spectral geometry of Riemannian submersions


Chapter 6: Lagrangian formalism on Grassmann manifolds

1 Introduction

2 Grassmann manifolds

3 The Lagrangian formalism on a Grassmann manifold

4 Lagrangian formalism on second order Grassmann bundles

5 Some applications


Chapter 7: Sobolev spaces on manifolds

1 Motivations

2 Sobolev spaces on manifolds: definition and first properties

3 Equality and density issues

4 Embedding theorems, Part I

5 Euclidean type inequalities

6 Embedding theorems, Part II

7 Embedding theorems, Part III

8 Compact embeddings

9 Best constants

10 Explicit sharp inequalities

11 The Cartan-Hadamard conjecture

Chapter 8: Harmonic maps Dedicated to the memory of James Eells


1 Harmonic functions on Euclidean spaces

2 Harmonic maps between Riemannian manifolds

3 Weakly harmonic maps and Sobolev spaces between manifolds

4 Regularity

5 Existence methods

6 Other analytical properties

7 Twistor theory and completely integrable systems

Chapter 9: Topology of differentiable mappings

1 Introduction

2 Manifolds and singularities

3 Milnor fibre

4 Monodromy

5 Stratifications of spaces

6 Stratified Morse theory

7 Rectified homotopical depth

8 Relative stratified Morse theory

9 Topology of images and multiple point spaces

10 The image computing spectral sequence

Chapter 10: Group actions and Hilbert’s fifth problem


1 Hilbert’s fifth problem

2 Lie groups and manifolds

3 Group actions

4 Cartan and proper actions of Lie groups

5 Non-paracompact Cartan G-manifolds

6 Homogeneous spaces of Lie groups

7 Twisted products

8 Slices

9 The strong Cr topologies, 1 ≤ r ≤ ∞, and the very-strong C∞ topology

10 Continuity of induced maps in the strong Cr topologies, 1 ≤ r < ∞, and in the very-strong C∞ topology

11 The product theorem

12 The equivariant glueing lemma

13 Whitney approximation

14 Haar integrals of Cs maps, 1 ≤ s ≤ ∞, and of real analytic maps

15 Continuity of the averaging maps in the strong Cr topologies, 1 ≤ r < ∞, and in the very-strong C∞ topology

16 Approximation of K-equivariant Cs maps, 1 ≤ s ≤ ∞, by K-equivariant real analytic maps, in the strong Cs topologies, 1 ≤ s < ∞, and in the very-strong C∞ topology

17 Approximation of Cs K-slices, 1 ≤ s ≤ ∞

18 Proof of the main theorem

Chapter 11: Exterior differential systems

1 Introduction

2 Exterior differential systems

3 Basic existence theorems for integral manifolds of C∞ systems

4 Involutive analytic systems and the Cartan-Kähler Theorem

5 Prolongation and the Cartan-Kuranishi Theorem

6 A Cartan-Kähler Theorem for C∞ Pfaffian systems

7 Characteristic cohomology

8 Topological obstructions

9 Applications to second-order scalar hyperbolic partial differential equations in the plane

10 Some applications to differential geometry

Chapter 12: Weil bundles as generalized jet spaces


1 Weil algebras

2 Weil bundles

3 On the geometry of TA-prolongations

4 Fiber product preserving bundle functors

5 Some applications

Chapter 13: Distributions, vector distributions, and immersions of manifolds in Euclidean spaces

1 Introduction

2 Distributions on Euclidean spaces

3 Distributions and related concepts on manifolds

4 Vector distributions or plane fields on manifolds

5 Immersions and embeddings of manifolds in Euclidean spaces

Chapter 14: Geometry of differential equations


1 Geometry of jet spaces

2 Algebra of differential operators

3 Formal theory of PDEs

4 Local and global aspects

Chapter 15: Global variational theory in fibred spaces

1 Introduction

2 Prolongations of fibred manifolds

3 Differential forms on prolongations of fibred manifolds

4 Lagrange structures

5 The structure of the Euler-Lagrange mapping

6 Invariant variational principles

7 Remarks


Chapter 16: Second Order Ordinary Differential Equations in Jet Bundles and the Inverse Problem of the Calculus of Variations

1 Introduction

2 Second-order differential equations on fibred manifolds

3 Variational structures in the theory of sodes

4 Symmetries and first integrals

5 Geometry of regular sodes on × TM

6 The inverse problem for semisprays


Chapter 17: Elements of noncommutative geometry

1 Introduction

2 Algebras instead of spaces

3 Modules as bundles

4 Homology and cohomology

5 The Chern characters

6 Connections and gauge transformations

7 Noncommutative manifolds

8 Toric noncommutative manifolds

9 The spectral geometry of the quantum group SUq(2)

Chapter 18: De Rham cohomology


1 De Rham complex

2 Integration and de Rham cohomology. De Rham currents. Harmonic forms [20]

3 Generalizations of the de Rham complex

4 Equivariant de Rham cohomology

5 Complexes of differential forms associated to differential geometric structures

Chapter 19: Topology of manifolds with corners

1 Introduction

2 Quadrants

3 Differentiation theories

4 Manifolds with corners

5 Manifolds with generalized boundary

Chapter 20: Jet manifolds and natural bundles

1 Introduction

2 Jets

3 Differential equations

4 The calculus of variations

5 Natural bundles

Chapter 21: Some aspects of differential theories Respectfully dedicated to the memory of Serge Lang


1 Background

2 Calculus in topological vector spaces and beyond

3 The Chern – Rund derivative

Chapter 22: Variational sequences


1 Preliminaries

2 Contact forms

3 Variational bicomplex and variational sequence

4 C-spectral sequence and variational sequence

5 Finite order variational sequence

6 Special topics

7 Notes on the development of the subject

Appendix: splitting the exterior algebra

Chapter 23: The Oka-Grauert-Gromov principle for holomorphic bundles


1 Stein manifolds and Stein spaces

2 Oka’s theorem

3 Grauert’s Oka principle

4 Gromov’s Oka principle

5 The case of Riemann surfaces

6 Complete intersections

7 Embedding dimensions of Stein spaces

8 Oka principle with growth condition

9 Oka’s principle and the Moving Lemma in hyperbolic geometry

10 The algebraic version of Oka’s principle





Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands

Linacre House, Jordan Hill, Oxford OX2 8DP, UK

First edition 2008

Copyright © 2008 Elsevier B.V. All rights reserved

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher

Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; email: permissions@elsevier.com. Alternatively you can submit your request online by visiting the Elsevier web site at http://www.elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material


No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-444-52833-9

For information on all Elsevier publications visit our website at books.elsevier.com

Printed and bound in The Netherlands

08 09 10 11 12      10 9 8 7 6 5 4 3 2 1


The group of topics known broadly as ‘Global Analysis’ has developed considerably over the past twenty years, to such an extent that workers in one area may sometimes be unaware of relevant results from an adjacent area. The many variations in notation and terminology add to the difficulty of comparing one branch of the subject with another.

Our purpose in preparing this Handbook has been to try to overcome these difficulties by presenting a collection of articles which, together, give an overall survey of the subject. We have been guided in this task by the MSC2000 classification, and so the scope of the Handbook may be described by saying that it covers the 58-XX part of the classification: ranging from the structure of manifolds, through the vast area of partial differential equations, to particular topics with their own distinctive flavour such as holomorphic bundles, harmonic maps, variational calculus and non-commutative geometry. The coverage is not complete, but we hope that it is sufficiently broad to provide a useful reference for researchers throughout global analysis, and that it will also be of benefit to mathematical physicists and to PhD and post-doctoral students in both areas.

The main work involved in the preparation of the Handbook has, of course, been that of the authors of the articles, who have carried out their task with skill and professionalism. Our debt to them is immediate and obvious. Some other potential authors have, for personal reasons, been unable to offer contributions to the Handbook, but we hope that those omissions will not detract too much from its value. The editors also wish to acknowledge the assistance of Petr Volný in the formatting of the LATEX manuscripts, and of Andy Deelen, Kristi Green and Simon Pepping at Elsevier for their help and advice during the preparation of the book. In addition, we should like to record our particular thanks to Arjen Sevenster from Elsevier, who commissioned the project and gave us support and encouragement during its development.

We should finally like to acknowledge the support of the Czech Science Foundation (grant 201/06/0922) and the Czech Ministry of Education, Youth, and Sports (grant MSM 6198959214) for our work on this project.

Demeter Krupka and David Saunders,     Palacký University, Olomouc.

Global aspects of Finsler geometry¹

Tadashi Aikou,     Department of Mathematics and Computer Science, Kagoshima University, 1-2-35 Korimoto, Kagoshima, 890 Japan. E-mail address: aikou@sci.kagoshima-u.ac.jp

Laszlo Kozma,     Institute of Mathematics, University of Debrecen, H-4010 Debrecen, P O Box 12, Hungary. E-mail address: kozma@math.unideb.hu


1 Finsler metrics and connections

2 Geodesics in Finsler manifolds

3 Comparison theorems: Cartan-Hadamard theorem, Bonnet-Myers theorem, Laplacian and volume comparison

4 Rigidity theorems: Finsler manifolds of scalar curvature and locally symmetric Finsler metrics

5 Closed geodesics on Finsler manifolds, sphere theorem and the Gauss-Bonnet formula

1 Finsler metrics and connections

1.1 Finsler metrics

Let π : TM M be the tangent bundle of a connected smooth manifold M of dim M = n. We denote by v = (x,y) the points in TM if y ∈ π−1(x) = TxM. We denote by z(M) the zero section of TM, and by TMx the slit tangent bundle TM\z(M). We introduce a coordinate system on TM as follows. Let U M be an open set with local coordinate (x¹, …, xn). By setting v = Σyi (∂/∂xi)x for every v ∈ π−1(U), we introduce a local coordinate (x, y) = (x¹, …, xn, y¹, …, yn) on π−1(U).

Definition 1.1

A function F : TM is called a Finsler metric on M if

(1) F(x, y) ≥ 0, and F(x, y) = 0 if and only if y = 0,

(2) F(x, λy) = λF(x, y: λ>0},

(3) F(x, y) is smooth on TMx, the out-side of the zero section,

(4) G = F²/2 is strictly convex on each tangent space TxM, that is, the Hessian (Gij) defined by


is positive-definite,

are satisfied. The pair (M, F) is called a Finsler manifold.

We note that the last condition in this definition is equivalent to the convexity of the unit ball Bx = {y TxM | F(x,y) ≤ 1}.

y of each y TxM y = F(x,y), and the length s(t) of a smooth curve c(t) = (x¹(t),…, xn(t)) is defined by


Example 1.2

(Funk metric) Let g be a Riemannian metric on M. We define α : TM . Since α is convex, there exists a 1-from β such that β(v) ≤ α(v). The function F = α + β defines a convex Finsler metric on M so-called Randers metric. We shall review a typical example of Randers metric (see n be an n-dimensional Euclidean space with the standard coordinate (x¹, …, xn), and B the unit ball centered the origin: B = {x n | φ(xx ²>0}. The Riemannian metric gH defined by

is called the Hilbert metric on B. We define a 1-form β by

H of β with respect to gH β(xH x <1, and thus the function F is a Finsler metric called the Funk metric on B. We note that the relation between gH and F is given by

for all v TM.

For the differential π* of the submersion π : TMx → M, the vertical subbundle V of T(TMx) is defined by V = ker π*, and V is locally spanned by {∂/∂y¹, …, ∂/∂yn} on each π−1 (U). Then it induces the exact sequence



is the pull-back bundle π*TM.

Since the natural local frame field {∂/∂xi}i=1 …,n on each U on π−1(U), any section X is written in the form X = Σ(∂/∂xiXi for smooth functions Xi on each π−1 (U). Furthermore, since ker π* = V, the differential π* is given by π* = Σ(∂/∂xidxi.

We define a metric G by


for every section X = Σ(∂/∂xiXi and Y = Σ(∂/∂xjYj. We also set



for all sections X, Y, Z . It is trivial that C vanishes identically if and only if G is a Riemannian metric on M. This tensor field C is called the Cartan tensor field.

In the sequel, we use the notation Ais naturally identified with V ≅ ker π*, any section X is considered as a section of V. We denote by XV the section of V :

The following is trivial since (1.2) is exact.



+ ≅ {cI GL(n); c +} ⊂ GL(n) acts on the total space by multiplication

+. This action induces a canonical section ε of V defined by ε(v) = (v, v) for all v TM×. By the homogeneity of F, we have

, and we denote it by the same notation ∈, that is, ∈(x, y) = Σ(∂/∂xiyi. This section ∈ is called the tautological section of and


1.2 Ehresmann connection

For the submersion π : TM× → M, the vertical subbundle is defined by V = ker π*, while the horizontal subbundle H is defined by a subbundle H T(TM×) which is complementary to V. These subbundles give a smooth splitting


Although the vertical subbundle V is uniquely determined, the horizontal subbundle is not canonically determined. An Ehresmann connection of the submersion π : TM× → M is a selection of horizontal subbundles. In this report, we shall define this as follows.

Definition 1.3

An Ehresmann connection of the submersion π : TM× → M satisfying



If an Ehresmann connection θ is given, then a horizontal subbundle H is defined by H = ker θ. In this report, we shall assume that the subbundle H defined by θ is invariant by the action m•, that is, (mλ)*H = H o m+. This assumption is equivalent to


Remark 1.4: A linear connection of the tangent bundle TM is a selection of horizontal subbundles in GL(n)-invariant way. Thus, an Ehresmann connection θ in our sense is sometimes called a non-linear connection of TM.

In the sequel, we denote by Ak the space of smooth k-valued k-form on TM× respectively. We suppose that an Ehresmann connection θ is given. Then, the exterior differential d : Ak Ak+1 is decomposed into the form d = dH dV according to the decomposition (1.7), where dH is the differential along H and dV is the one along Vis also decomposed into the form D = DH DV.

Proposition 1.5

If an Ehresmann connection θ is given, then there exists a covariant exterior derivation D of satisfying


or equivalently


Proof: We define a covariant derivation D by DV = dV . It is easily shown that D = DH dV . Then we have

and, from (1.9) we obtain

Therefore we obtain (1.11).

is also naturally identified with the horizontal subbundle H, and any section X is considered as a section of H. We denote by XH the section of H :


denotes the horizontal lift of natural local frame field {∂/∂x¹, …, ∂/∂xn} with respect to the given Ehresmann connection θ. The set {dx¹, …, dxn} is the dual basis of H*. For the two bundle morphism π* and θ from T(TM, we have

Proposition 1.6

The bundle morphisms π* and θ satisfy




for every X A).

1.3 Chern connection

If a Finsler metric F is given on TM, then there exists a natural metric G defined by (1.3). Then we shall introduce a covariant derivation ∇ which satisfies some natural axioms.



With respect to the splitting (1.7), ∇ is also decomposed into the form ∇ = ∇H V.

Definition 1.7

([10]) The Chern connection on (M, Funiquely determined from the following conditions.

(1) ∇ is symmetric:


where we considered π* = Σ(∂/∂xidxi .

(2) ∇ is almost G-compatible:


Remark 1.8: In the case of C = 0, the metric F is the norm function of a Riemannian metric g, and the Chern connection ∇ is given by ∇ = π* ∇M for the Levi-Civita connection ∇M of (M, g). The Chern connection is also called the Rund connection of (M, F) (cf. [3], [12] [32]).

We can easily show that θ defined by (1.14) is invariant by the natural action m+. In local coordinate, θ is given by

is the connection forms of ∇ with respect to {∂/∂x¹, …, ∂/∂xn}. The set {θi, …, θnis the dual basis of V* defined by θ.

Then the covariant derivative ∇ is also decomposed into the form ∇ = ∇H Vrespectively. The covariant derivative ∇G of the metric G is decomposed into the form ∇G = ∇H G + ∇V G, and thus the assumption :


By the definition (1.4) of Cartan tensor field C, we have


for all X, Y, Z A).

On the other hand, and

Therefore we obtain

Proposition 1.9

Let be the Ehresmann connection of π : TM× → M defined by (1.14) for the Chern connection of (M, F). Then we have


Since the condition . Then the condition (1.17) is written as dHG tωG Gare given by


where (Gij) denotes the inverse of (Gij).

1.4 Parallel translation

Let (M, F) be a Finsler manifold with the Chern connection ∇. For a non-vanishing vector field v = Σvi(x)(∂/∂xi) on M, we define its covariant derivative ∇v be the natural lift of v . The covariant derivative ∇v :


If v satisfies ∇v = 0, then v is said to be parallel with respect to ∇.

Let c = (x(t)) : I = [0,1] → M be a smooth curve, and v(t) be a non-vanishing vector field along cis said to be horizontal :


If v(t) satisfies this equation, v(t) is said to be parallel along c.

The system (1.22) has a unique solution vζ(t) which depends on the initial condition ζ = vζ(0) smoothly. From smooth dependence of solutions on ζ, the mapping Pc(t) : Tc(0)M Tc(t)M defined by Pc(t)(ζ) = (c(t), vζ(t)) is a diffeomorphism for every t I. Because of homogeneity of θ, if vζ(t) is a solution of (1.22), then λvζ(t) is also a solution satisfying λvζ (0) = λζ, and thus the uniqueness of solutions implies that vλζ (t) = λvζ(tof a curve c starting at ζ ∈ Tc(0)M , and Pc(t) also satisfies the homogeneity Pc(t)(λζ) = λPc(t)(ζ) for all λ>0 and ζ ∈ Tc(0)M. The family Pc = {Pc(t) : t I} is called the parallel translation along c with respect to ∇.

The tangent space TxM at every point x M x = F(x, •). If we put Pc(t)(ζ) = (c(t), vζ(t)) for any point ζ in Tc(0)Mvζ(tof the vector field vζ(t) along c(t) is given by F(c(t), vζ(t)). Then, because of Proposition 1.9 we have

Hence the parallel translation Pc Pc(tc(tc(0).

Proposition 1.10

The parallel translation Pc along any curve c = c(t) on M is a norm-preserving map between the tangential normed-spaces.

The parallel translation Pc is said to be isometry if it satisfies


for all ζ, η ∈ TpM. The parallel translation Pc along a curve c = c(t) is norm-preserving, but not isometry in general. It is trivial that, if Pc is a linear mapping, then Pc is an isometry. In a later section, we shall consider the case where every tangent spaces are isometric mutually as normed linear spaces.

We denote by Cp the set of all (piecewise) smooth curves c with starting point p = c(0) and ending point p = c(1). Then there exists a natural product ‘o’ in Cp. We also set Hp = {Pc(1) : TpM TpM | c Cp}. Defining PcPc2(1) = P(c1 · c2)(1), we can easily show that Hp is a group with the multiplication ‘o’. This group Hp is called the holonomy group with reference point p M. In general, since the parallel translation Pc is not linear between the fibres, Hp is not a Lie group.

1.5 Torsion

in the sequence , and the Chern connection ∇ makes π* parallel by the assumption (1.15). We shall calculate the covariant exterior derivative ∇π*:

Since V is integrable and π* satisfies (1.12), we have ∇π*(XV, YV, we have


Therefore we have

Proposition 1.11

The assumption (1.15) of the Chern connection is equivalent to




for all .

by (1.10). In Finsler geometry, the covariant exterior differential ∇θ plays an important role.

Definition 1.12

The torsion T of ∇ is defined by


We remark that the vertical part of T vanishes identically. In fact, we have

, since the vertical subbundle V is integrable. Similar computations lead us to

Proposition 1.13

The horizontal part THH and the mixed part THV are given by




for all respectively. In particular, the mixed part THV satisfies


where THH (X, Y) = T(XH, YH) and THV(X, Y) = T(XH, YV).

Remark 1.14: Using local coordinate (x¹, · · ·, xn), the torsion form T is given by


Case of: THH ≡ 0

The horizontal subbundle H is integrable if and only if [H, H] ⊂ H. Then, from (1.13), H is integrable if and only if θ([XH, YH]) = 0 for all X, Y A).

Definition 1.15

The integrability tensor of θ is defined by


From (1.27), we have THH = Θ, and thus THH = 0 means that the horizontal sub-bundle H is integrable. Therefore H on the total space TM× which is transversal to the fibres if THH ≡ 0.

Definition 1.16

A non-vanishing smooth section v : M TM× is said to be horizontal with respect to ∇ if it satisfies ∇v = v*∇ε = v*θ = 0.

For a horizontal section v, we have v*dV ≡ 0, and thus vdH = d v*.

From this, the integrability condition d(v*θ) = 0 for v to be horizontal is given by v*θ = 0. Hence, if we assume Θ ≡ 0, this assumption guarantees the existence of a horizontal section v(x) = (x, y(x)) satisfying v(x0) = ζ(≠ 0) for an arbitrary initial point ζ ∈ Tx0M. The n-dimensional submanifold of TM defined by y = y(xthrough the point (x0, ζ).

For the metric G , we define a Riemannian metric gv on M by gv = v*G for a horizontal section v. For the connection form ω of ∇, the form ωv := v*ω defines a connection ∇M on TM. Then we have

Proposition 1.17

Assume that THH ≡ 0. Let v : M TM× be a horizontal section, and gv the induced metric on M. The induced connection M on M is the Levi-Civita connection of (M, g), and v is parallel with respect to M.

Case of: THV ≡ 0 : Landsberg spaces

The metric G makes each fibre TxM a Riemannian space with the metric Gx. Since the parallel translation Pc along any curve c and its tangent vector field lies in horizontal space:

Pc is an isometry between the fibres if and only if


(cf. [23]). We have then

Proposition 1.18

The Lie derivative XHG is given by


for all .

Definition 1.19

A Finsler manifold (M, F) is said to be a Landsberg if THV ≡ 0.

From the relation (1.33), we can prove that (1.32) is equivalent to THV = 0.

Theorem 1.20

([22]) A Finsler manifold (M, F) is Landsberg if and only if the parallel translation Pc along any curve c is an isometry between the tangential Riemannian spaces.

From this theorem, we see that any parallel translation Pc Hp along c Cp is a isometry in the tangential Riemannian space TpM if (M, F) is a Landsberg space. Then it is shown that Hp is a Lie group, since the isometry group G of any Riemannian manifold is a Lie group.

Proposition 1.21

([26]) Suppose that (M, F) is a Landsberg space. Then the holonomy group Hp with reference point p M is a Lie group.

Remark 1.22

Each fibre TxM is a Riemannian manifold with the metric induced form G. Then, the condition THV = 0 means that each fibre TxM is totally geodesic in TM with some Sasakian-type metric (cf. [1]).

On the other hand, the volume form on each fibre TxM is induced from the n. We shall consider the case where each fibre TxM is minimal in TM. From (1.25), we have

. If we put (tr THV)(X) := trace{Y THV(X, Y)}, then, by direct computations, we can show the following

XH Π = 0 if and only if (tr THV)(X) = 0.

, we denote by φt the one parameter group of local transformation in TM induced from XH. For a compact subset K0 ⊂ Tx0M, x0 ∈ M, we set Kt = ϕt(K0) ⊂ Tϕt(x0)M. Then the volume V(Kt. Suppose that (M, F) is Landsberg space. Then by Theorem 1.20, each ϕt is isometry between the fibres, and we have

Hence, if (M, F) is Landsberg, the volume V(Kt) is constant(cf. [7]).

A Finsler manifold (M, F) satisfying tr THV = 0 is called a weak Landsberg space (cf. [44]). The condition tr THV = 0 means that each fibre TxM is is minimal submanifold in TM with some Sasakian-type metric (cf. [1]).

1.6 Curvature

An important quantity in geometry is the curvature which measures the flatness of the space.

Definition 1.23

The curvature R of ∇ is defined by


Similarly to the case of torsion T, we first remark that the vertical part of R vanishes identically. In fact, we obtain

, since the vertical subbundle V satisfies [V, V] ⊂ V. Hence the surviving part of R are the horizontal part RHH and the mixed part RHV:


The following is trivial from the definition.


From the assumption (1.15), the Ricci identity ∇² π* = R ∧ π* gives R ∧ π* = 0.



induce the following.

Proposition 1.24

For all , the curvatures RHH and RHV satisfy the following identities:




By the definition (1.26), the Ricci identity ∇²ε = Rε implies the relation T = Rε.

Proposition 1.25

The curvature R and the torsion T of the Chern connection satisfy the relations




for all .

The symmetry assumption (1.15) derives Proposition 1.24. The almost G-compatibility assumption (1.16) derives the followings. Firstly, concerning with RHH, we have

Proposition 1.26

The horizontal curvature RHH satisfies


for all X, Y, Z .

Secondary, concerning the mixed part RHV, we have

Proposition 1.27

The mixed part RHV satisfies the following identities.


for all sections X, Y, Z and W of .

As an application of Proposition 1.27, we obtain the following

Proposition 1.28

The mixed part RHV of the curvature R satisfies the following identity


for every X A).

Remark 1.29: The curvature form of ∇ is the 2-form on TM. By definition of R, we obtain the 1-st Bianchi identity Ω = dω + ω ∧ ω:


Differentiating this identity, we obtain the 2-nd Bianchi identity ∇Ω = 0:


In the case of Finsler geometry, this identity induces some complicated identities, since ∇ has non-zero torsions THH and THV. For example, if we calculate the horizontal 3-form of the left-hand-side of (1.44), we obtain

The other identities including the terms ∇H RHV, ∇V RHH and ∇V RHV, see the book [10] or [32].

Case of: RHH ≡ 0

We suppose RHH = 0. Then (1.38) gives θ = THH = 0, and thus TM× admits a horizontal section v : M TM×.

Proposition 1.30

If RHH = 0, the induced metric gv = v*G is a flat Riemannian metric on TM, and so M is locally Euclidean.

Case of: RHV = 0 : Berwald spaces

In this subsection, we shall consider the case of RHV = 0.

Definition 1.31

A Finsler manifold (M, F) is said to be Berwald if RHV = 0.

Because of (1.39), a Berwald space is a special class of Landsberg spaces. If (M, F) is Berwald, the Chern connection ∇ is linear, that is, there exists a symmetric linear connection ∇ on TM such that ∇ = π*∇M.

If (M, F) is a Berwald space, then Szabó’s theorem [49] showed that we can find a Riemannian metric g on M which is compatible with ∇M. We show the outline of the proof of this fact. For this, we define a isometric group G of F. For an arbitrary point x M, we set G = {g GL(ngy y , ∀y TxMand the homogeneity of F, we can prove that G is a compact Lie group [21]. Then we define an inner product (·, ·)x on TxM by

for an arbitrary inner product (·, ·) on TxM and a bi-invariant Haar measure dg on G. By definition, it is trivial that this inner product 〈·, ·〉x is (G-invariant.

On the other hand, the holonomy group Hx of ∇M with the reference point x is a subgroup of G, since ∇M preserves F invariant. Hence 〈·, ·〉x is also Hx-invariant. Thus we can extend the inner product 〈·, ·〉x to a Riemannian metric g on TM by the help of the parallel displacement with respect to ∇M. It is trivial that ∇M is compatible with respect to this metric g. Hence we have

Theorem 1.32

([49]) Suppose that a Finsler manifold (M, F) is a Berwald space. Then there exists a Riemannian metric g on M such that the Chern connection of(M, F) is given by ∇ = π*∇M for the Levi-Civita connection M of(M,g).

The curvature is related with parallel translation. The following theorem due to [21] characterizes Berwald spaces in terms of parallel translations.

Theorem 1.33

([21]) A Finsler manifold (M, F) is Berwald if and only if the parallel translation Pc along any curve c is an isometry between the tangential normed spaces.

Proof: We suppose that the parallel translation Pc along any curve c = c(t) in M is an isometry between the tangential normed spaces. Then, by the Mazur-Ulam’s theorem (cf. [33] and [50]), the mapping Pc is linear and there exists a GL(nsatisfying

. From (1.22) we get

for all curves c(tare linear with respect to the fibre coordinate (y¹, · · ·, yn, which shows RHV = 0.

Conversely we suppose that (M, Fare linear in the fibre coordinate (y¹, · · ·, yn), the solutions yζ(t) of (1.22) are linear in ζ. Hence the parallel translation Pc is linear, and from (1.23), we have

for all ζ, η ∈ Tc(0) M. Hence Pc is an isometry.

Example 1.34

Let F = α + β be a Randers metric on M. Then, by the well-known theorem due to [25], (M, F) is Berwald if and only if β is parallel 1-form on the base Riemannian manifold (M, g).

Remark 1.35: From (1.39), if RHV vanishes identically, then THV also vanishes. Hence the class of Landsberg spaces contains the class of Berwald spaces. There exists a lot of example of Berwald spaces. However it is still an open problem to find an example of non-Berwald Landsberg space.

Case of: RHH = RHV = 0 : Locally Minkowski spaces

In this section, we shall be concerned with flat Finsler manifolds.

Definition 1.36

A Finsler manifold (M, F) is said to be locally Minkowski if there exists a local coordinate system on M with respect to which the function F is independent of the base point x M.

We have the following well-known theorem (cf. [32]).

Theorem 1.37

A Finsler manifold (M, F) is locally Minkowski if and only if the Chern connection is flat.

Proof: We suppose that ∇ is flat, that is, RHH = RHV = 0. Then, the Chern connection ∇ is induced from a flat connection ∇M on TM. Hence there exists an open cover U of M and local frame fields (e1, · · ·, en) on U ∈ U such that ∇ej = 0. This condition is equivalent to the existence of the change of frames

on each U satisfying dA + ωA = 0. With respect to such a local frame field eof ∇ vanishes on Uand (1.19) imply the independence of F on the base point x Uthe inverse of the matrix A, the condition above is equivalent to

Since ∇ is symmetric, there exist some functions wi(x, and the local frame ej is given by

Hence there exists a local coordinate system {U, (wi)} on M with respect to which the function F is independent on the base point x U.

Conversely we assume that F is independent of the base point x. By definition, the metric tensor Gij is also independent of the base point xand and thus the Chern connection ∇ is flat.

Example 1.38

Let F = α + β be a Randers metric on M. Then (M, F) is locally Minkowski if and only if (M, F) is Berwald and the base Riemannian manifold (M, g) is flat ([25]).

This example is true for any locally Minkowski space.

Proposition 1.39

A Finsler manifold (M, F) is locally Minkowski if and only if (M, F) is Berwald and its associated Riemannian metric is flat.

1.7 Flag curvature

Let X be a tangent vector at x M. We may consider X . Then the 2-plane F(X) spanned by X and ε is called the flag with the flagpole ε. For the curvature tensor R of ∇, the sectional curvature

is called the flag curvature of the flag F(X), and denoted by K. From (1.42), we have G(RHV(X, ε) ε, X, and so the flag curvature K is given by


The flag curvature K depends on X and the point (x, y) ∈ TM× : K = K(x, y, X). If the flag curvature K is independent of X at every point (x, y) ∈ TM×, the space is said to be of scalar flag curvature K(x, y). A Finsler manifold (M, F) is said to be of constant flag curvature if K is constant. For Finsler manifolds of constant flag curvature, see Chapter 12 in [10]. The proof of the following theorem is found in [32].

Theorem 1.40

(Schur’s lemma) Let (M, F) be a Finsler manifold of scalar flag curvature K = K(x, y). If K is a function of position x M alone, then (M, F) is of constant flag curvature provided dim M ≥ 3.

Example 1.41

Let F be the Funk metric on the unit ball B defined in Example 1.1. It is well-known that F has negative constant flag curvature K = −1 (see [37] or [14]).

2 Geodesics in Finsler manifolds

Let γ : I = [0,1] → M be a smooth curve. Since the symmetry condition F(x, y) = F(x, −y) is not assumed, the orientation of curves is essential, that is, if a curve γ is given, then we always assume that γ is oriented by the parameter t. A smooth curve γ = γ(t) is said to be regular for every t I.

Let Γ(p, q) be the set of all regular oriented curves with the initial point p = γ(0) and the terminal point q = γ(1). Then we define a functional LF : Γ(p, qby

Since F satisfies the homogeneity condition, this definition is well-defined. For an ordered pair (p, q) ∈ M × M, the distance function dF(p, q) is defined by dF(p, q) = infc LF(γ), where infinimum is taken over of all oriented (piecewise) smooth curves from p to q. In general, since the symmetry condition is not assumed, the distance function df does not satisfy the symmetric property dF(p, q) = dF(q, p). However, the distance dF satisfies the following conditions:

(1) dF(p, q) ≥ 0,

(2) dF(p, q) = 0 if and only if p = q,

(3) dF(p, q) ≤ dF(p, r) + dF(r, q).

The metric topology of M is defined by the sets B(p, δ) = {q M | dF(p, q)<δ}, and the metric topology of a connected Finsler manifold (M, F) coincides with the manifold topology of M.

2.1 Geodesics in Finsler manifolds

The canonical lift For a vector field X(t) along γ, we consider X(t) as a section of TM . Then we use the notation ∇tX :


, since X is a vector field along the curve γ(t) in the base manifold M. The definition (2.1) and the metrical condition (1.14) imply


for all vector fields X(t) and Y(t) along γ. Hence we have

Proposition 2.1

If ∇tX = ∇tY = 0, then the inner product G(X, Y) is constant along γ.

Let (M, F) be a Finsler manifold with the Chern connection ∇.

Definition 2.2

A regular oriented curve γ : I M is said to be a path .

The length s , and the function s(t) is an increasing function of the parameter t. If the parameter t is positively proportional to s, then t is said to be normal.

Definition 2.3

Let (M, F) be a Finsler manifold. A path in M with a normal parameter is called a is a geodesic in (M, F).

From (2.1), a regular oriented curve γ(t) = (xi(t)) with normal parameter t :


is satisfied. From the metrical condition (2.2), if γ = γ(t) with normal parameter t has a constant norm and γ has constant speed. In the sequel, we always assume that the parameter of a geodesic to be normal otherwise stated.

Let γX : I M be a geodesic with initial point x = γX, where the parameter t is, of course, is normal. We shall define the exponential map exp by exp (x, X) = γX(1) if X = 0 and exp (x, 0) = x. The restriction of exp to D ∩ TxM is denoted by expx. The restricted exponential map expp maps the rays through the origin 0 ∈ TxM to the unique geodesics through the point x in sufficiently small Bx(r) = {X TxM X <r}.

The exponential map exp is defined on an open neighborhood D of the zero section z(M) of TM, and exp is C∞-class away from z(M). Furthermore exp is C¹-class at z(M), and its derivative at z(M) is the identity map. By a result due to Akbar-Zadeh, the map exp is C²-class at z(M) if and only if (M, F) is a Berwald space (see [10]).

For each X TxM, the radial geodesic γX is given by γX(t) = expx(tX) for all t I such that either side is defined. This geodesic segment γX , is constant along γX

2.2 The first variation of arc length and geodesics

In this section, we shall show the first variation formula in Finsler manifolds. For this end, we introduce some definitions.

Let γ = γ(t) ∈ Γ(p, q. Then a variation of γ is a family {γs} of oriented curves γs(t) parameterized by s ∈ (-ε, ε) such that γ0(t) = γ(t) for all t I. A variation Γγ is said to be proper if it fixes the end points, that is, γs(0) = p and γs(1) = q. We suppose that the map Γγ : (-ε, ε) × I M defined by Γγ(s,t) = γs(t) is smooth. (For the variational problem of arc length, it is enough to assume that Γγ is piecewise differentiable with respect to the parameter t (cf. [32], Chapter VII), however, we shall assume the smoothness of Γγ for the simplicity of discussions.)

By the assumption, the map Γ satisfies Γγ(0, t) = γ(t), p = Γγ(s, 0) and q = Γγ(s, 1). Setting s = constant for each s ∈ (-ε, ε), the parameterized curve γs : I M defined by γs(t) = Γγ(s, t) is called a s-curve, while the parameterized curve γt(s) = Γγ(s,t) is a t-curve which is transversal curve to γ. In local coordinates, we set Γγ(s,t) = (x¹(s,t), · · ·, xn(s, t)). We denote by S = ∂γt/∂s and T = ∂γs/∂t the tangent vector fields of t-and s-curves respectively:

In particular, the vector field V(t) along γ defined by

is called the variational field induced from Γγ. If Γγ is proper, that is, Γγ satisfies γs (0) = γ(0) = p and γs (1) = γ(1) = q for all s ∈ (-ε, ε), then the variational field V is proper, that is, V satisfies V(0) = V(1) = 0.

at least one point on γ. Let V = V(t) be any vector filed along a regular oriented curve γ = γ(t). Then there exists a variation Γγ which induces V as its variational field. In fact, if we take Γγ(s, t) = exp(sV(t)), then Γγ : (-ε, ε) × I M is a variation of γ with variational field V.

Lemma 2.4

Let V be any vector field along γ. Then V is a variational field of some variation Γγ of γ. If V is proper, then V is the variational field induced from a some proper variation Γγ.

of s-curve γs:

Lemma 2.5

Let Γγ : (-ε, ε) × I M be a variation. Then we have



Let Γγ be a proper variation of a regular oriented curve γ ∈ Γ(p, q). We compute the first variation of length functional LFs). Since

, we have

Furthermore, (1.17) and (2.4) imply


. Consequently we have

which gives

Evaluating s derives the following:

Proposition 2.6

(First Variation Formula) Let γ : I M be a regular oriented curve, and Γγ a proper variation of γ. Then


where V is the variational field of Γγ.

A regular oriented curve γ is said to be a stationary point of the functional LF if (dLFs)/ds)s=0 = 0 for any proper variation Γγ. If a regular oriented curve γ : I M is a geodesic, then γ satisfies (2.3), and thus γ is a stationary point of LF from (2.6).

Conversely we suppose that γ is a stationary point of the functional LF. Since the condition (dLFs)/ds)sfor a smooth function ϕ satisfying ϕ(0) = ϕ(1) = 0 and ϕ>0 elsewhere. Then, since V is proper and from (2.6), we have

on I.

Proposition 2.7

A regular oriented curve in a Finsler manifold (M, F) is a stationary point of the functional LF if and only if γ is a geodesic from p to q.

A curve γ from p = γ(0) to q = γ(1) is said to be minimizing if dF(p, q) = LF(γ). Since minimizing curve is a stational curve, we have

Theorem 2.8

Every minimizing curve in (M, F) is a geodesic if γ is regular.

The converse of this theorem is also true.

Theorem 2.9

Every geodesic in a Finsler manifold (M, F) is locally minimizing.

This theorem is proved by using the Gauss lemma. We define the geodesic ball x(r) centered at x M of radius r x(r) = exp (Bx(r)) for the tangential ball Bx(r) = {ζ ∈ TxM <r}. Let Sx(r) = {X TxM X = r} be the tangent sphere. Then the set Sx(r) = exp (Sx(r)) is called the geodesic sphere at x of radius r. Then the Gauss lemma is stated as follows.

Lemma 2.10

(The Gauss Lemma) The radial geodesic γX is orthogonal to the geodesic sphere Sx(r) at x M.

For the proof of Theorem 2.9, we need more technical preliminaries, and thus we omit it here. For the complete proof, see [10] or [13].

2.3 Euler-Lagrange equation

From Proposition 2.8, a geodesic in (M, F) is characterized as the critical points of length functional LF. In general, the equation which characterizes the critical points of a functional on Γ(p, q) is called the Euler-Lagrange equation of LF. For details, refer to the book [4] or [32].

We consider an arbitrary proper variation Γγ : γs(t) = γ(t) + sX(t) of a smooth curve γ(t) with fixed end points, that is, X(0) = X. The Taylor extension gives

and this extension implies

Because of

we have

Hence, γ(t) is a critical point of LF if and only if


This equation is the Euler-Lagrange equation of the functional LF. The quantity Ei is written as

where we put

Putting Gi = Σ Gim Gm and yi = dxi/dt, the Euler-Lagrange equation Ei(γ) = 0 implies the equation of the geodesic as follows:

for the function p = dlogF/dt, where we note that t is an arbitrary parameter of γ. In particular, if we take Finslerian arc length as the parameter t of γ, that is, dt = F(x, dx), we obtain the equation of geodesic as follows:


Example 2.11

(Geodesics of Randers metrics) Let F = α+β be a Randers metric, where α² = Σ gij (x)yiyj and β = Σ bi(x)yithe Chiristoffel symbol of the base Riemannian manifold (M, g). Let t be the Finslerian arc length.


Then, by direct computations, we see that the functions Gi in (2.8) are given by

Hence the equation (2.8) with the Finslerian arc length parameter t is given by the following complicated form:

If we take the Riemannian arc length u as the parameter, that is, du = α(x, dx), we have dt = dubi(x)dxi, and so

Then the equation above is reduced to the following form:

In particular, if the 1-form β = Σ bi(x)dxi implies

and so any geodesic in (M, F) coincides with one in the base Riemannian manifold (M, g). Consequently it is shown that, if the 1-form β is closed, then (M, F) is projectively equivalent to the base Riemannian manifold (M, g).

Example 2.12

(Geodesics of Funk metric) Let F n stated in Example 1.1. In this case, the 1-form β is exact form, and so any geodesic in (B, F) is given by the one in the Hilbert’s space (B, gH). We shall show the equation of geodesic parameterized by its Finslerian arc length t. From Example 2.1, it is given by

Since the Hilbert’s metric gH = Σ gij (x)dxi dxj is given by

is given by

Then, because of

we obtain

Hence the equation (2.8) is given by

Furthermore, from

, we obtain

Consequently, the equation of geodesics in (B, F) is given by the following simple form:

The solution of this differential equation with initial conditions A = (a¹, · · ·, an) = xis given by xi = λi(1 – et)+ai. Therefore any geodesic in (B, F) is given by a line in B.

Let P the point (λ¹+a¹, · · ·, λn+an) = x(∞) on the boundary ∂B, and B a point on the line. We denote by AP (resp. BP) the Euclidean distance between the points A and P (resp. B and P). Because of

which implies that the Funk’s distance df is given by

Since the Hilbert’s distance dH is given by dH(A, B) = [dF(A, B)+dF(B, A)]/2, the distance dH is given by

2.4 The Jacobi fields and conjugate points

A variation Γγ = Γγ(s,t) of a geodesic γ is said to be a geodesic variation if each s-curve γs is also a geodesic. Since each s-curve γs .

Let X be a vector field along γs. Then, since [S, T] = 0, we have


along γs. From this equation, we get the so-called the Jacobi equation.

Proposition 2.13

(The Jacobi Equation) Let γ be a geodesic, and V the variational field of a geodesic variation Γγ of γ in a Finsler manifold (M, F). Then V satisfies


Definition 2.14

Let (M, F) be a Finsler manifold. The differential equation (2.10) is called the Jacobi equation. A vector field J along a geodesic satisfying (2.10):

is called a Jacobi field in (M, F).

By definition, the variational field V of a geodesic variation of a geodesic γ is a Jacobi field. Conversely, every Jacobi field along a geodesic γ is the variational field of some geodesic variation of γ. The differential equation (2.10) is linear and of second order, we have 2n linearly independent solution. Therefore, along any geodesic γ, the set of Jacobi field is a 2n-dimensional vector space.

Let γ ∈ Γ(p, q) be a geodesic segment in M. Then q is said to be conjugate along γ if there exists a Jacobi field J(≠ 0) along γ such that J vanishes at p and q.

For X TpM, we set q = expp X. For an arbitrary Y TX (TpM), we shall compute the differential (expp)*Y at X:

To compute (expp)*, we define a geodesic variation Γγ of γX by Γγ(s, t) = expp t(X+sY). The variational field J = ∂ Γγ/∂s is a Jacobi field along γX, and we have J(1) = (expp)*Y. The conjugate points are the image of the singularities by the exponential mapping.

Proposition 2.15

Let γX(t) = expp(tX) (t ε I) be the radial geodesic for X ε TxM. Then expp is a local diffeomorphism if and only if q = expp X is not conjugate to p along γX.

2.5 The second variational formula and index form

Let γ : I M be a geodesic with unit speed. We shall compute the second variation of the length functional LF. We shall compute

Differentiating with respect to s, we have

From (2.2) and (2.4), we get


Consequently we have

along γsand V(0) = V(1) = 0 imply

we have


and ∇t V⊥ = ∇t V – ∇t = (∇t V)⊥. Hence we have

from from (1.36). Hence we get

Consequently, we obtain the second variation formula of LF.

Proposition 2.16

(Second Variation Formula) Let γ : I M be any geodesic with unit speed, and Γγ a proper variation of γ, and V its variation field. Then


where V⊥ is the normal part of V.

Since (1.36) implies

, we obtain

for the flag curvature K. Hence the second variation formula (2.12) has the form


Therefore we have

Proposition 2.17

Let (M, F) be a Finsler manifold with non-positive flag curvature K. Then, the second variation of any geodesic satisfies

We define the index form on a Finsler manifold (M, F). Let γ be a unit speed geodesic in (M, F). We set


for normal proper vector fields X, Y along γ. The index form I is a symmetric bi-linear form on the space of normal proper vector fields. In fact, the Bianchi identity (1.36) implies

Since the last term on the left hand side vanishes from (1.6) and (1.42), we have

along γ. Thus I is a symmetric bi-linear form: I(X, Y) = I(Y, X).

Since (1.35) induces

along γ, if X and Y are proper, we have

which implies


By the definition of I and (2.11), the second variation of Lp of unit speed geodesic is given by I(X, X), and it can be thought as the Hessian of the length functional LF. Thus, if γ is minimizing, then I(X, X) ≥ 0 for any proper normal vector field X along γ. The following is a generalization of the well-known theorem in Riemannian geometry which shows that no geodesics is minimizing past its first conjugate point (e.g., Theorem 10.15 in [30]).

Theorem 2.18

If γ ε Γ(p,